Artificial Neural Networks (ANN) have potential of the parallel processing by the integrated circuit tech-nology. Recently, over one million gates are available by the latest FPGA (Field Programmable Gate Array). However, the sum-of-product circuit used for evaluating inputs of neurons is complex and not effective for hardware implementation by FPGAs. In this paper, an improved calculation algorithm of the neuron model is proposed, which is based on the multi-dimensional binary search. Since the search does not need the sum-of-product circuit, the designed neuron circuit is small and fast. It is suitable for hardware implementation.

抄録全体を表示

The theory of descriptor system (DS) has been studied intensively in recent years, but little attention was paid to study the model following control system (MFCS) for DS. In this paper, a design method of nonlinear model following control system with disturbances is proposed. This method can handle with differential and algebraic equations simultaneously. Consequently it is useful for nonlinear electric circuits and multi-linkage mechanism system and so on. As regular transformation is not used in control input design, the system structure and the physical properties of system variables can be preserved. The norm conditions of nonlinear part are necessary for proof of bound on inner states, but those conditions are not strong constraint because many practical nonlinear systems satisfy. In addition, the parameters of these norm con-straints are just for the purpose of proving the bound on inner states and have no influence on design of control system. The bound on inner states for the closed loop system can be proved with using the theory of poles and zeros of DS, characteristic of nonlinear function and positive real transfer function. Finally two numerical simulations with disturbances verify the availability of this method.

抄録全体を表示

A well known control system which can reduce the adverse effect of a input disturbance is a disturbance Observer. However, in many case of mechanical systems a system disturbance which does not satisfy match-ing condition may be impressed. Therefore, it may be difficult to reduce the adverse effect of the disturbance by the use of the traditional disturbance observer.
In this paper, as a prerequisite for the system disturbance which can be observed, a construction of control system for the disturbance rejection using a dual observer is proposed. This method is based on the zero output by disturbance localization problem. This problem may be solved by dividing the state space into observable subspace and unobservable subspace. As compared with H_∞ controller based on perfect observation, the usefulness of the proposed disturbance rejection control system was shown by numerical simulations for 2-mass spring system.

抄録全体を表示

The authors propose a method, applied to automatic facial caricaturing, for extracting faces and facial parts from a grayscale image; an outline of the method and experimental results are presented in this paper. The irises of the eyes were first extracted using a Hough transformation for circles. Based on the location of the extracted irises, the eyebrows, nose, and mouth were extracted by a simple method based on the profile. An experiment using 96 grayscale facial images showed recognition rates of at least 95%. To extract the shapes of the facial parts, a generalized Hough transformation was applied to the edge patterns in the respective facial parts' regions. The parameter space of this Hough transformation was characterized by the dominant coefficients of the K-L expansion applied to the sampled data of the facial part. An experiment attempting extraction of the eye boundary was performed, and it was confirmed that the eye boundaries can be successfully extracted with a limited number of boundary feature points.

抄録全体を表示

Market mechanism solves the product distribution problem by allocating the scheduled resources according to market prices. Conventional market-oriented program never deals with due date utility in its nature, although that is one of the major factors in practical trading process. In this paper a new concept based on the market-oriented programming with two types of due date utilities of agents is proposed so as to facilitate a Pareto-optimal product allocation amongst various agents in resource allocation problems. Simulation results indicate the effectiveness of the proposed algorithm and detect further research subjects.

抄録全体を表示

Previous work by the authors has proposed a banknote recognition system using a neural network (NN) to develop new types of banking machines such as banknote readers and sorters. Up to now, the feasibility and effectiveness of our research for neuro-recognition system is shown using the large quantity of worldwide banknotes in the commercial market. However, in this neuro-recognition system, we had to transport the banknotes data from the banking machines to a personal computer (PC) in case of learning. After learning on the PC, neural parameters such as weights and masks are downloaded to the neural board. In this way, we have coped with the new banknotes or fluctuation of banknotes in the commercial market. In this paper, we implement the neuro-learning algorithm to this board and extend its specification in order to realize the intelligent machines, which can be tuned by itself according to the banknotes. Furthermore, we newly define specification information, which absorbs difference of some applications, for wide use of this board. Then we advance new neuro-recognition system to palm prints recognition. We construct an experimental system for palm prints recognition using this neuro board, image scanner and PC. Finally, we show the possibility of expansion for this board, concretely.

抄録全体を表示

The dynamics of power systems contain the nonlinearities such as trigonometric functions and the param-eter uncertainties of reactances of transmission lines caused by short circuit faults. Wang et al. proposed an adaptive nonlinear controller for a single generator connected through two parallel transmission lines to an infinite bus. They emplyed the feedback linearization technique and the linear adaptive control technique. Though they used the power angle, the relative speed of the generator and the active electrical power as feed-back signals, the power angle is not usually available. Therefore, we design an adaptive nonlinear controller without the power angle feedback-path for Wang's power systems model. We employed the adaptive control design method with a Riccati equation. This method allows us to design the parameter adjustment law only using the power angle of the generator.

抄録全体を表示

A position and orientation measuring system for an indoor mobile robot is proposed based on “RECS” (Robot Environment Compromise System) concept. RECS concept makes it possible to develop the multipurpose robot by setting the environment so as that the robot can act skillfully. As the setting of the environment, a simple and cheep landmark is developed by using a commercial retro-reflect type reflection sheet and it is attached to the ceiling of a room. This landmark is based on white color and is not conspicuous to a human being at usual state, and is illuminated from the robot in order to be conspicuous only when the measuring procedure is needed. Considering economical efficiency and unobtrusiveness, this landmark can be employed at many points in the room. The robot can obtain landmark's image in a CCD camera and can know the absolute position of the landmark by reading the barcode. A calculating method of robot's position and orientation is newly proposed, which compensates two landmarks' images by information of robot's inclinations and uses them. Static and dynamic measuring accuracy of this system is experimentally inspected and it is less than 10mm and ideg in the neighborhood of two landmarks. By selecting appropriate two landmarks from many of them, this accuracy is thought to be kept in the wide area of the room.

抄録全体を表示

In this paper, a new method for faster neural networks learning is proposed. Characteristic of our method is that let neural networks have functions of synaptic weights instead of synaptic weights in order to improve the sensitivity of the criterion functions with respect to the synaptic weights. By constructing the functions of synaptic weights appropriately, the learning process can be significantly improved. By a simulation study of learning of controller parameters for a nonlinear crane system control, it is clarified that the speed of learning by the proposed method is much faster than that of the conventional method with moment.

抄録全体を表示

This paper proposes two methods for connected component (CC) labeling with run-based way in binary images. In one method of the two, called method A, a binary image is transformed into run representation, and then each run is labeled in raster scan order. Labeled runs are transformed into a label image, finally. Labeling in raster scan order produces equivalent labels. We use the union-find algorithm with path compression and weight balancing for equivalent label maintenance. In another method, called method B, transforming of a binary image into runs and labeling of runs proceed simultaneously in raster scan order. The method B has a merit of saving memory space because only current two lines are required as run information. In the general case where both of input binary images and output label images are two-dimensional array representation, both of proposed methods are faster than the pixelwise CC labeling standard method and rnn-based label nrnnacation methods.

抄録全体を表示

The purpose of this paper is to theoretically identify the condition for an optimal control problem in this type of system with transaction lost in stationary. We propose the transportation process based on the counting process in input/output process. Accordingly, we make clear this stationary optimal condition for this type of system. We also provide an optimal control solution for the maximum revenue by using a value function in transportation process.

抄録全体を表示

A new self-creating type self-organizing neural network is proposed. It is well known that the columnar structure in the human brain plays an important role in visual information processing. In the columnar network, cells which represent similar features are gathered nearby. This structure is useful for robust in-formation processing. Realization of the structure in the computational model is effective for intelligent information processing. The key concept of the proposed model lies in the self-creation of new nodes with weight duplication. The daughter node is created based on a self-organizing neural network. The mother node has refractory period just after the creation. The weights of the daughter node and those of her mother node become similar after the refractory period. A hierarchical learning leads the mother-daughter relation-ship to represent the similar features of the columnar network. This results in self-creation of the nodes which represent similar features without dead nodes. Simulations are carried out to show feasibility of the proposed model.

抄録全体を表示

First, we explain the technique to produce the face expression change using the basic picture such as the expressionless face and the smiling automatically by the interpolation ability of the Kohonen neural networks (KNN) which is known as the competition networks. The three dimensional shape and the color information of the facial surface which is obtained by the range finder are used for the learning of the KNN. Next, as the method of searching the optimal path of the face expression change, DP matching is used. It uses the weight which was generated by the learning of KNN for the distance calculation of the DP matching. Lastly, we examine about the difference with the face expression which the photograph was actually taken and the face expression which was produced by KNN.

抄録全体を表示

A gene expression system n-BDD (n-output Binary Decision Diagram) was proposed in order to express and optimize agent's behavior ⁽⁶⁾. Although the system is suitable for behavior models of agents, it does not include crossover. This paper proposes a crossover operation using Bryant's Apply operation⁽¹²⁾ and a simulation framework of quasi-ecosystem using n-BDD. The operation makes an n-BDD probabilistically inherit two functions expressed by two n-BDDs. In an experiment the proposed method had more than 40% high fitness than the conventional method. Using this method we model an environment where carnivores and herbivores are co-evolved, we find that a food chain relation in the system using the method with crossover continued a longer time than in the conventional method.

抄録全体を表示

The estimation of power system transient stability is more difficult and of great importance because the power system becomes larger and larger. Therefore, the fast stability estimation to contingency faults is demanded in the power system operation. In this paper, the neural network technique is applied for the fast classification of system dynamic security status. A suitable topology for the neural network is developed, and the appropriate training method and the input and output signals are selected. The procedure developed is successfully applied to the IEEE 10-generator 39-bus test system. The data obtained by the energy functions are used for training the neural network.

抄録全体を表示

This paper derives an empirical formula of an attenuation constant inside small cross-section tunnels (2m²-13m²) which are used as headraces in hydroelectric power stations. E lectric feild strength/distance inside the tunnel was measured (1.9GHz-18GHz), and the obtained data was submitted to statistical analysis.
The attenuation constant obtained from linear regression analysis was larger than that reported by researchers.
With the attenuation constant as a criterion variable, multiple regression analysis was conducted. Examining the correlation and regression coefficients gave two parameters with highly partial F values. One is the fundamental attenuation constant term of (λ²/r³), the same as the previously reported parameter, and the other is the additional loss term of (λ^0.28/e^0.66r) We proposed.
The parameters yielded an empirical formula which estimates the attenuation constant in the waveband (0.006m≤λ≤1.5m) broader than before. [4175•(λ²/r³)+64.4•(.λ^0.28/e^0.66r)] The attenuation constant using the empirical formula agreed with that obtained from experiments, confirming the effectiveness of our proposed empirical formula.
It was concluded that the attenuation constant in the Giga-hertz range includes additional loss, resulting in the analysis value larger than the previous value.

抄録全体を表示

So-called geriatric diseases such as diabetes and hyperlipemia increase with changing life style of the modern time. In order to undergo a medical checkup of the private health condition, simple and convenient instru-ment detecting biochemical information is a thing of daily necessity. Several methods of measuring glucose and cholesterol concentration in the blood have recently been developed by means of optical device or enzyme electrode. For the daily use of the instrument, some requirements like high accuracy, fast response, small quantity of sampling, simple handling as well as low price must be considered. In this study, we proposed an optical biochemical sensor using three-layer slab type of thin-film glass waveguide to measure glucose and cholesterol concentration and investigated basic characteristics of the sensor. Principle of measurement is based on the absorption of evanescent wave of the guided laser light. For attaining high sensitivity and fast response, direct enzymatic reaction of the sample evolved with the reagent on the sensing layer of waveguide. Transient response of the output signal can be represented by double exponential curve fitting, in which fast time constant of the relaxation response shows dose dependency of the sample concentration. From these results, it is concluded that high sensitivity. fast response and small quantity of the measurement can be obtained for biochemical samples by using optical thin film waveguide.

抄録全体を表示

We have investigated sea clutter amplitudes by a millimeter wave radar using an Akaike Information Cri-terion (AIC). It is shown that sea clutter obeys a log-Weibull distribution for all data and data within the beam width of an antenna. To suppress such sea clutter, we have applied the Constant False Alarm Rate (CFAR) procedure by the nonlinear transformation to the real data.

抄録全体を表示

In this paper, we propose a system of evolving virtual machine structure in three-dimensional space. It is easy for users to design the structure of machines in the system by use of Computer Graphics (CG). The system can acquire suitable machine structures for given virtual world by evolutionary approach. In the proposed system, first, the virtual environment in which virtual machines move is set. Next, all types of the machine parts data are inputted to the system. A machine structure consists of parts data and the information on the combination. One machine structure is transformed to one gene. The first population of genes is made of combination of a few parts selected at random. Offspring are generated by evolutionary techniques: selection, mutation, and crossover. The machine structure is adapted to the virtual world by repetition of evolution. We assumed a flat plane with various obstacles as a virtual environment. It is confirmed that various structures could be obtained through computer experiments.

抄録全体を表示

Fast and accurate prediction of three-dimensional (3D) power distribution is essential in a boiling water reactor (BWR). The prediction method of 3D power distribution in BWR is developed using the neural network. Application of the neural network starts with selecting the learning algorithm. In the proposed method, we use the learning algorithms based on a class of Quasi-Newton optimization techniques called Self-Scaling Variable Metric (SSVM) methods. Prediction studies were done for a core of actual BWR plant with octant symmetry. Compared to classical Quasi-Newton methods, it is shown that the SSVM method reduces the number of iterations in the learning mode. The results of prediction demonstrate that the neural network can predict 3D power distribution of BWR reasonably well. The proposed method will be very useful for BWR loading pattern optimization problems where 3D power distribution for a huge number of loading patterns (LPs) must be performed.

抄録全体を表示

We propose a learning automaton computing method (LACM) with a genetic algorithm (GA) for the multi-variable function optimization problems and complete a distributed parallel implementation. The simulation results show that LACM has better computing accuracy and higher convergence property than other tradi-tional methods.

抄録全体を表示